Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Alex Gavryushkin arXiv:1504.01180v2 [math.LO] 23 Nov 2015 DECIDABLE MODELS OF SMALL THEORIES Department of Computer Science, The University of Auckland, Private Bag 92019, Auckland, New Zealand email: [email protected] Abstract. Many counterexamples are known in the class of small theories due to Goncharov [3] and Millar [5]. The prime model of a decidable small theory is not necessarily decidable. The saturated model of a hereditarily decidable small theory is not necessarily decidable. A homogeneous model with uniformly decidable type spectra is not necessarily decidable. In this paper, I consider the questions of what model theoretic properties are sufficient for the existence of such counterexamples. I introduce a subclass of the class of small theories, which I call AL theories, show the absence of Goncharov-Millar counterexamples in this class, and isolate a model theoretic property that implies the existence of such anomalies among computable models. I consider only countable structures of countable languages. And only small theories. Definition 1. A first-order theory T is small if the set of finite first-order types of T without parameters, S(T ), is countable. Throughout the paper all theories are assumed to be small. Fact 1. Every small theory has a prime model and a saturated model. These models are unique up to isomorphism. Fact 2. If p is a type of a small theory T and A |= p(ā) then the theory T h(A, ā) has a prime model (Aā , c̄). The isomorphism type of the structure Aā does not depend on the choice of A and ā, it only depends on the 2000 Mathematical Subject Classification. 03C57, 03D75. Key words and phrases. theories, types, decidable theories, AL theories, computable and decidable models. This work was partially funded by the subsidy allocated to Kazan Federal University for the state assignment in the sphere of scientific activities. 1 2 A. GAVRYUSHKIN type p. Since we consider structures up to isomorphism, denote the structure by Ap . Definition 2. Call the structure Ap from Fact 2 p-prime, or almost prime if the type is not specified. The set AP T of all almost prime models of a theory T is preordered by the relation of elementary embeddability. The preorder induces a partial order on the factor-set AP T / ∼, where A ∼ B ⇔ (A B & B A), in the natural way. Note that (AP T / ∼, ) has a unique least element—the prime model of T . Definition 3. We call the partial order (AP T / ∼, ) the fundamental order of the theory T . The notion of fundamental order is also known as Rudin–Keisler order [6, 7]. In this paper, I follow the terminology of Lascar–Poizat [4] and Baldwin–Berman [1], who introduced and studied fundamental orders of theories in general. To illustrate the definition, let us consider the following simple though important examples of fundamental orders. Example 1. The following theories T are AL theories in the sense of Definition 5 below. (1) A saturated structure is almost prime if and only if the structure is ℵ0 -categorical. For the theory T of such a structure, AP T is the one-element partial ordering. (2) If a theory T is ℵ1 - but not ℵ0 -categorical then AP T ∼ = ω. (3) If a complete theory T has finitely many countable models then AP T has a greatest element. Proof. (1) is obvious. For (2) we note that the saturated model of such a theory is not almost prime and the rest of models are almost prime. (3) A type p of a theory T is called powerful if every model of T realising p realises every type of T as well. If T is a complete theory with finitely many, but more than one, countable models then it has a non-principal powerful type p. A p-prime structure gives the greatest element of AP T . Proposition 1 ([6]). If Ap ∼ Aq but Ap ∼ 6= Aq , then there is a structure A such that A ∼ Ap but A is not almost prime. Proof. Form an elementary chain A0 ⊆ A1 ⊆ . . . where An ∼ = Ap if n is S ∼ even and An = Aq if n is odd. Put A = An . n∈ω DECIDABLE MODELS OF SMALL THEORIES 3 The structure A can be presented as a union of an elementary chain of isomorphic almost prime structures, but A itself is not almost prime. Following Sudoplatov [6, 7], I shall call such a structure limit: Definition 4. A structure is p-limit if it is a union of an elementary chain of p-prime structures but it itself is not p-prime. A structure is limit if it is a union of an elementary chain of isomorphic almost prime structures but it itself is not almost prime. Definition 5. A complete small theory T is an AL theory if every countable non-saturated model of T is either almost prime or limit. Definition 5 can be rewritten syntactically [6]. Example 1 above gives some well-known examples of AL theories. Note that a saturated structure is limit if and only if its theory has a non-principal powerful type, i. e. AP T has a maximal element. Let LS T denote a set of all limit models of an AL theory T . The structure of the spectrum of models of the theory T is determined by a pre-ordering AP T and a function λT : AP T → 2LS T mapping a p-prime structure to the set of all p-limit structures [6, 7]. Think of λT as of a S λT (M). disjoint union of bipartite graphs. LS T = M ∈AP T Question 1. How distinct are the class of AL theories and the class of small theories? Naturally, the class of AL theories is a proper subclass of the class of small theories. The following definition isolates a model theoretic property of small theories that might be violated in the class of AL theories. Sudoplatov proved in [6] that if we relax Definition 5 using Definition 6 then we get the class of all small theories. Definition 6 ([7]). A structure is weakly limit if it is the union of an elementary chain of almost prime structures. For the sake of completeness, I include the proof of Sudoplatov’s lemma [6] that every countable model of a small theory is either almost prime or weakly limit. Lemma 1 (S. Sudoplatov [6]). Every countable model of a small theory is either almost prime or weakly limit. Proof. Let A be a countable model of a small theory T . Let a0 , a1 , . . . be an enumeration of the domain of A. Consider the type tp(a0 ) of a0 in A. Since T is small, there is a prime model (A0 , a0 ) of tp(a0 ) in 4 A. GAVRYUSHKIN the signature enriched by one constant. We can choose A0 to be an elementary submodel of A. If A0 is isomorphic to A, we are done and A is almost prime. If not, take the first element b1 in our enumeration a0 , a1 , . . . such that b1 ∈ A \ A0 . Consider a prime model (A1 , a0 , b1 ) of tp(a0 , b1 ). Again, if A1 is isomorphic to A then A is almost prime. If not, we continue the process. If the process terminates after finitely many steps, the model A is prime over (a0 , b1 , . . . , bn ) for some n and hence is almost prime. If not, we obtain an elementary chain A0 A1 . . . of almost prime structures. Since a0 , a1 , . . . is an enumeration of all elements of A, we have that ∪i Ai = A and A is weakly limit. The following general theorem shows that the class of AL theories is rather different from the class of all small theories if we study decidable models. This theorem is the key property to the absence of GoncharovMillar counterexamples in the class of AL theories. Theorem 1. Let T be an AL theory. Then the set of decidable almost prime models of T forms an ideal in the fundamental order of T . Proof. Let us first note the following property of decidable models of an AL theory, which is interesting on its own. Lemma 2. Let T be an AL theory and A its non-saturated decidable model. Then there exist a type p and a decidable p-prime model Ap such that the type spectra of the structures A and Ap coincide. Proof. If A is almost prime, there is nothing to prove. Suppose A is plimit for some p. Since A is decidable, the set of types realised in A is uniformly computable. This gives a uniformly computable enumeration of principal types realised in (Aā , c̄), where ā is a realisation of p in A. Hence, (Aā , c̄) is decidable and so is Aā . It remains to note that the type spectra of A and Aā coincide. I now prove that the set of decidable models of an AL theory is directed upwards. Lemma 3. Let T be an AL theory and A and B be its decidable models. Then there exists a decidable model C such that C A and C B. Proof. Using Lemma 2 above, one can assume that A is a p-prime and B is a q-prime structure for some decidable types p and q. Take C to be a decidable structure that realises both p and q. Such a structure exists because otherwise the theory T would have more than countably many types. DECIDABLE MODELS OF SMALL THEORIES 5 To finish the proof of the theorem, we need to show that the set of decidable almost prime models of T is closed downwards. Let Ap be a decidable almost prime model of T and Aq be an almost prime model of T such that Aq Ap . We want to show that Aq is decidable. Since T is AL, there are only finitely many almost prime structures with pairwise distinct type spectra A0 = Aq , A1 , . . . , An = Ap such that A0 A1 . . . An . Using Millar’s type omitting theorem [5] (see also [3]), we can realise all the types that are realised in Aq and omit all the types that are not realised in Aq , in a decidable structure B. That is, the structure B has the same type spectra as Aq and is decidable. We apply Lemma 2 to prove that Aq is itself decidable. Since the set of decidable almost prime models of the theory T is directed upwards and closed downwards, it forms an ideal. Corollary 1. If an AL theory T is decidable then T has a decidable prime model. Proof. Since the theory T is decidable, it has a decidable model A. If the model A is saturated then the prime model of T is decidable, see [3]. If the model A is not saturated then it is p-prime or p-limit and we apply Theorem 1 to prove that the prime model of T is decidable. Note that Corollary 1 does not hold in the class of all small theories. In [2], we construct a decidable small theory T whose all types are decidable yet whose prime model is not decidable. Millar [5] was the first to construct such an example but his construction uses an infinite language in an essential way, while our structure is a graph. Since the prime model of a theory with decidable saturated model is decidable, we have an example of a decidable small theory whose saturated model is not decidable. Although the class of AL theories and the class of small theories have similar model theoretic properties, they are significantly different from the computability point of view. Corollary 1 along with the following corollary demonstrate this difference. Corollary 2. If T is a decidable AL theory whose types are all decidable, then every almost prime model of T is decidable. Proof. Let Ap be an almost prime model. Since p is a decidable type, it can be realised is a decidable model. Applying Theorem 1 we get that Ap is decidable. In the light of Theorem 1, a natural question to ask would be what ideals can be formed by decidable models? The following theorem of 6 A. GAVRYUSHKIN ours answers this question in full for the class of theories with lattice-like fundamental orders. The question remains open in the general case of an arbitrary fundamental order. Let AP D T be the ideal formed by decidable models of an AL theory T . Theorem 2 (A. Gavryushkin and B. Khoussainov [2]). Let L be a finite lattice and L′ be its ideal. Then there exists an AL theory T such that: (1) The fundamental order of T is L, that is, (AP T / ∼, ) ∼ = L. ∼ (2) The spectra of decibel models of T is L′ , that is, AP D L′ . = T Computable models behave rather differently than decidable ones. I conclude with a theorem that shows that computable models do not form an ideal in the class of all models of an AL theory. Theorem 3 (A. Gavryushkin and B. Khoussainov [2]). For every finite lattice L, there exists a theory T of finite signature with countably many models such that: (1) The fundamental order of T without the least element is isomorphic to L. (2) For all p ∈ L, the class of models corresponding to p contains infinitely many models of which exactly one is computable. References [1] J. Baldwin and J. Berman, Concrete representations of lattices and the fundamental order, Classification Theory, 24–31, 1987. [2] A. Gavryushkin and B. Khoussainov. On decidable and computable models of theories. Springer LNCS, Vol. 7921, 200–209 (2013). [3] S.S. Goncharov and Yu.L. Ershov. Constructive Models. Springer (2000). [4] D. Lascar, B. Poizat, An introduction to forking, The Journal of Symbolic Logic, 44, 3, 330–350, 1979. [5] T. Millar. Foundations of recursive model theory. Annals of Mathematical Logic, 13, 1, 45–72 (1978). [6] S.V. Sudoplatov. Complete theories with finitely many countable models I. Algebra and Logic, 43, 1, 62–69 (2004). [7] S.V. Sudoplatov. Hypergraphs of prime models and distributions of countable models of small theories. Journal of Mathematical Sciences, 169, 5, 680–695 (2010).